Can machine learning predict onset of mental depression?

Machine learning algorithms are being used to predict onset of depression

Machine learning algorithms are being used to predict onset of depression


Ellie, touted as the world’s first AI-psychologist, has sparked significant debates on the extent machine learning can alter the mental health landscape. For the unfamiliar, Ellie is a visual diagnostic tool that can read as many as 60 non-verbal human behavioural cues such as eye-gaze, face tilt, etc. per second. With this data, Ellie is envisaging it can identify and detect early warning signs of depressions. Ellie has been jointly developed by researchers at the DARPA and USC’s Institute of Creative Technologies. A major thrust for its development was from the intended objective to stem the rise of suicide cases among security forces members.

Is Ellie working? Initial reports suggests it is. Security forces members, who have been seeking counselling help, prefer to interact with Ellie over a human counterpart. Skeptics might argue that the initial euphoria must be driven by inquisitiveness around the AI’s human like capabilities rather than competence. But in the longer scheme of things, it is undeniable that with time, the science of psychology is bound to be automated and machine learnt.

Ellie isn’t the only potentially breakthrough product we are looking at. Apart from body language, other individual traits can also be potentially leveraged to predict onset of mental illness. Researchers at the Columbia Medical Center and the New York State Psychiatric Institute used machine-learning algorithms to digitally analyze speech patterns in written transcripts. This model accurately predicted onset of mental illness among 34 youths, who participated in the experiment. How does the system work? By using NLP techniques, it detects anomalies in word patterns within patient interviews transcripts. It then uses algorithms to correlate absence of expected pattern of words and phrases with the likely risk of psychosis.

Behavioral data from smartphones can too play a vital role in diagnosis of mental illness. In Dartmouth, Andrew Campbell, a computer scientist, is experimenting with smartphone data from students to predict their mental health and behavioral patterns. Why was he doing these experiments? Andrew hypothesized that the cumulative effect of a number of behavioral factors play a significant role in the student’s ability to channelize his energies for success – in this case, academic performance. Using a mobile app to measure vital parameters such as sleep cycles, chats, location, etc. and running simple predictive algorithms, Andrew concluded that engaging in conversations and physically activities curbed the potential of academic failure. So are we talking about a future where student mental health will be monitored by tracking his daily activity? Perhaps yes. And given the fact that recent isolated shooting incidents have suggested deteriorating mental health of a student can be lethal for the community as well, it might be not be an entirely bad idea to be considered. Obviously, issues related to an individual’s privacy needs to be addressed first.

A far more comprehensive and commercialized expansion of the above model is GingerIO. GingerIO uses smartphone to track user activity and consequently, employs analytical engines to predict risk of mental health issues. Planning to expand its horizon across the health spectrum, its first commercial products focus on depression, co-morbid depression, anxiety disorders, bipolar disorder, and schizophrenia. GingerIO has already tied up with more than 20 US hospitals, mostly mental health institutions.

Talking of smartphones, one mustn’t overlook the ground-breaking work being done by the Copenhagen based startup Monsenso. Monsenso is a mobile health app that provides patients with an improved ecosystem (dashboard) to adhere to treatment plan and analyze their mental health records, and care providers with a tool to manage as well as analyze the mental health treatment performance of all the patients. Recent research shows that the adherence rate of self-assessments when using the Monsenso smartphone app is 88%, a remarkable number. Monsenso is also helping care providers with machine learning tools to peep into and understand patient behavior in a smarter way.
In addition to these, there have been several other academic research initiatives that are making significant path-breaking headway in using machine learning to manage mental health issues.

For e.g., researchers at MIT are trying to predict if existing diseases adversely affect brain anatomy. For this, the team is using genetic, demographic and clinical data to build its machine learning algorithm. Analyzing MRI data from patients with neuro-degenerative diseases, the team is successfully improving the prediction probability of the system.

Cambridge Cognition, Harimata
MRI data is being used elsewhere as well. Cambridge Cognition, developers of neuro-psychological tests for clinical practice and academia, is using machine learning to analyze brain scan MRI data and predict the onset of dementia. Similarly, Poland based Harimata is using a child’s performance on video games to predict and diagnose cases of autism. The child plays a game on a tablet and the app gathers thousands of data points from the sensors, such as the gyroscope and accelerometer used for analysis by the predictive model.

Researchers at JHU Human Language Technology Center for Excellence, led by Glen Coppersmith, are conducting NLP analysis on Twitter feed, along with metadata information (location of tweet, interaction degree) to predict risk of mental health issues. Their prediction model to identify people diagnosed with depression, bipolar disorder, PTSD or seasonal effective disorder is at works. Another such effort is being run jointly by Dartmouth University and the U.S. Department of Veteran Affairs. Focusing exclusively on war veterans, The Durkheim Project (project name) analyses social media accounts and mobile phones of veterans to track their mental health. Durkheim project is not the only such effort out there. David Cooper, a Mobile Health Program psychologist at the National Center for Telehealth and Technology (T2) has been working with the Department of Defense to help security forces community with mental health care tools. Cooper has already develop mobile apps such as Cooper has helped develop apps like PTSD Coach, Mood Tracker, Breathe to Relax and Virtual Hope Box.

A WHO estimate has predicted that, by 2030, depression will be the leading cause of global disease burden. Hopefully, some of these initiatives will go a long way in timely arresting the spread of depression.

Anubhav, a data scientist, writes about new developments and future trends in the machine learning and data analytics domain.
He can be reached at
Follow him on Twitter at:

Leave a Reply